Deep Learning Framework for Forest Fire and Smoke Detection Using DC-GAN and YOLO26n Algorithm

Abstract

Forest fires are considered as devastating consequences on ecosystems, biodiversity, human populations, and enormous amounts of carbon dioxide emissions, conducive significantly to climate change, it is most destructive natural calamities worldwide. Some of the main approaches for reducing the damage caused by wildfires are monitoring and early detection thus, are subjects of active research in artificial intelligence and computer vision. In this work, suggest a deep learning (DL) method combining the DCGAN (Deep Convolutional Generative Adversarial Network) based data augmentation with YOLOv26n approach for early fire and smoke detection. The CLAHE (Contrast Limited Adaptive Histogram Equalization) technique is used for pre-processing dataset images because CLAHE improves the mean average precision and recall by visually enhancing the image in degraded fire scene conditions. Compare the novelty model YOLO26n with YOLOv12N, YOLOv8, YOLOv11 with D- fire dataset which takes care of fast classification and detection of fires. This approach for classification facilitates spatial localization along with categorization of fires in the scene. Evaluation metrics such as 92.7% of accuracy, 90.3% of precision, 89.1% of recall, and 0.15 of Loss, 90% mAP@0.5 and mAP50:95 of 47.6% etc were achieved.